Modified neutron-netns-cleanup-cron.py script.

We needed to terminate qdhcp namespaces which aren't satisfied hosts in neutron db.
It happens after people interrupt neutron agents work and after restart agents, they can start ns on different hosts,
but nothing tracks previous hosts. Previous version of the script checked only IPs inside ns but didn't validate hosts.

Change-Id: I9968f627ce3ab1596711fe9e8d3345d0a5fc42c8

neutron/Chart.yaml

@@ -14,7 +14,7 @@ apiVersion: v1
 appVersion: v1.0.0
 description: OpenStack-Helm Neutron
 name: neutron
-version: 0.2.18
+version: 0.2.19
 home: https://docs.openstack.org/neutron/latest/
 icon: https://www.openstack.org/themes/openstack/images/project-mascots/Neutron/OpenStack_Project_Neutron_vertical.png
 sources:

neutron/templates/bin/_neutron-netns-cleanup-cron.py.tpl

@@ -1,18 +1,135 @@
 #!/usr/bin/env python
 
 import sys
+import os
 import time
-
+import socket
+from neutron.common import config
 from oslo_config import cfg
+from neutron.agent.linux import dhcp
+from neutron.agent.l3 import namespaces
+from neutron.agent.l3 import dvr_snat_ns
+from neutron.agent.l3 import dvr_fip_ns
+from neutron.cmd.netns_cleanup import setup_conf
+from neutron.cmd.netns_cleanup import destroy_namespace
+from neutron.cmd.netns_cleanup import eligible_for_deletion
+from neutron.conf.agent import common as agent_config
+from neutron.agent.linux import ip_lib
+from keystoneauth1.identity import v3
+from keystoneauth1 import session
+from neutronclient.neutron import client as neutron_client
+NS_PREFIXES = {'l3': [namespaces.NS_PREFIX, dvr_snat_ns.SNAT_NS_PREFIX,
+                      dvr_fip_ns.FIP_NS_PREFIX]}
+DHCP_NS_PREFIX = dhcp.NS_PREFIX
 
-from neutron.cmd.netns_cleanup import main
+def get_neutron_creds():
+    opts = {'auth_url': os.getenv('OS_AUTH_URL', 'https://keystone-api.openstack.svc.cluster.local:5000/v3'),
+        'password': os.getenv('OS_PASSWORD','nopassword'),
+        'project_domain_name': os.getenv('OS_PROJECT_DOMAIN_NAME', 'default'),
+        'project_name': os.getenv('OS_PROJECT_NAME', 'admin'),
+        'user_domain_name': os.getenv('OS_USER_DOMAIN_NAME', 'default'),
+        'username': os.getenv('OS_USERNAME', 'admin'),
+        'cafile' : os.getenv('OS_CACERT','/var/lib/neutron/openstack-helm/openstack-helm.crt'),
+        'insecure' : os.getenv('NEUTRON_CLEANUP_INSECURE', 'true'),
+        'debug': os.getenv('NEUTRON_CLEANUP_DEBUG', 'false'),
+        'wait': os.getenv('NEUTRON_CLEANUP_TIMEOUT', '1800')}
+    return opts
+
+def net_list(neutron_get):
+    hosts = dict()
+    net_list = neutron_get.list_networks()
+    if net_list['networks']:
+        for item in net_list['networks']:
+            net_id=item['id']
+            dhcp_agents = neutron_get.list_dhcp_agent_hosting_networks(net_id)['agents']
+            agents = list()
+            if dhcp_agents:
+                 for agent in dhcp_agents:
+                     agents.append(agent['host'].split('.')[0])
+            hosts[net_id] = agents
+    return hosts
+
+def sort_ns(all_ns, dhcp_prefix):
+    dhcp_ns = list()
+    not_dhcp_ns = list()
+    for ns in  all_ns:
+        if ns[:len(dhcp_prefix)] == dhcp_prefix:
+            dhcp_ns.append(ns)
+        else:
+            not_dhcp_ns.append(ns)
+    return dhcp_ns, not_dhcp_ns
+
+def del_bad_dhcp(dhcp_ns, dhcp_hosts, conf, dhcp_prefix, debug):
+    for ns in dhcp_ns:
+        cut_ns_name = ns[len(dhcp_prefix):]
+        if cut_ns_name in dhcp_hosts:
+            if hostname not in dhcp_hosts[cut_ns_name]:
+                destroy_namespace(conf, ns, conf.force)
+                if debug:
+                    sys.stdout.write("DEBUG: {} host {} deleted {} because host wrong\n"
+                                     .format(sys.argv[0], hostname, ns))
+            else:
+                if debug:
+                    sys.stdout.write("DEBUG: {} host {} {} looks ok\n"
+                                     .format(sys.argv[0], hostname, ns))
+        else:
+            destroy_namespace(conf, ns, conf.force)
+            if debug:
+                sys.stdout.write("DEBUG: {} host {} deleted {} because no related network found\n"
+                                 .format(sys.argv[0], hostname, ns))
+
+def del_bad_not_dhcp(not_dhcp_ns, conf, debug):
+    for ns in not_dhcp_ns:
+        if eligible_for_deletion(conf, ns, conf.force):
+            destroy_namespace(conf, ns, conf.force)
+            if debug:
+                sys.stdout.write("DEBUG: {} host {} deleted {} because no IP addr\n"
+                                 .format(sys.argv[0], hostname, ns))
 
 if __name__ == "__main__":
+
+    conf = setup_conf()
+    cfg.CONF(sys.argv[1:])
+    opts = get_neutron_creds()
+    debug = False
+    verify= False
+    if opts.pop('debug') in ('true', '1', 'True'):
+        debug = True
+    insecure = opts.pop('insecure')
+    cafile = opts.pop('cafile')
+    if insecure in ('false', '0', 'False'):
+        verify = cafile
+    timeout = int(opts.pop('wait'))
+    conf()
+    config.setup_logging()
+    agent_config.setup_privsep()
+    auth = v3.Password(**opts)
+    hostname = socket.gethostname().split('.')[0]
+
     while True:
         try:
-            main()
-            # Sleep for 12 hours
-            time.sleep(43200)
+            all_ns = ip_lib.list_network_namespaces()
+            sess = session.Session(auth=auth, verify=verify)
+            neutron_get = neutron_client.Client('2.0', session=sess)
+            dhcp_hosts = net_list(neutron_get)
+            if all_ns:
+                dhcp_ns, not_dhcp_ns = sort_ns(all_ns, DHCP_NS_PREFIX)
+                if dhcp_ns:
+                    del_bad_dhcp(dhcp_ns, dhcp_hosts, conf, DHCP_NS_PREFIX, debug)
+                else:
+                    if debug:
+                        sys.stdout.write("DEBUG: {} host {} no dhcp ns found\n"
+                                         .format(sys.argv[0], hostname))
+                if not_dhcp_ns:
+                    del_bad_not_dhcp(not_dhcp_ns, conf, debug)
+                else:
+                    if debug:
+                        sys.stdout.write("DEBUG: {} host {} no not_dhcp ns found\n"
+                                         .format(sys.argv[0], hostname))
+            else:
+                if debug:
+                    sys.stdout.write("DEBUG: {} host {} no ns found at all\n"
+                                     .format(sys.argv[0], hostname))
         except Exception as ex:
             sys.stderr.write(
                 "Cleaning network namespaces caught an exception %s"
@@ -24,3 +141,4 @@ if __name__ == "__main__":
             time.sleep(30)
         finally:
             cfg.CONF.clear()
+        time.sleep(timeout)

neutron/templates/daemonset-netns-cleanup-cron.yaml

@@ -76,6 +76,10 @@ spec:
             - /etc/neutron/dhcp_agent.ini
             - --config-file
             - /etc/neutron/l3_agent.ini
+          env:
+{{- with $env := dict "ksUserSecret" $envAll.Values.secrets.identity.admin "useCA" false }}
+{{- include "helm-toolkit.snippets.keystone_openrc_env_vars" $env | indent 12 }}
+{{- end }}
           volumeMounts:
             - name: pod-tmp
               mountPath: /tmp

releasenotes/notes/neutron.yaml

@@ -32,4 +32,5 @@ neutron:
   - 0.2.16 Remove usage of six
   - 0.2.17 Migrated PodDisruptionBudget resource to policy/v1 API version
   - 0.2.18 Updated naming for subchart compatibility
+  - 0.2.19 Added qdhcp NS host validation for deleting wrong namespaces.
 ...